🔍 Read the full analysis: What The School AI Pact And California Laws Say About Human Judgment on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Microsoft says that from November 1, 2026, it is making a school AI agreement available to US school districts, with provisions restricting student-data use, monitoring and consequential decisions without human review. The agreement is contractual, not a new law, and applies to education-designed products; California and New York have adopted separate legal limits that take effect on different schedules.
Microsoft says it is offering a school AI contract to every U.S. school district starting November 1, 2026, setting out restrictions on student-data use, monitoring and decisions made without meaningful human review. The agreement is not a new law: it is a proposed contractual standard, while recently signed California and New York laws impose separate requirements in workplaces, health care and chatbot services for minors.
The 31-page Memorandum of Agreement was announced September 9 by the American Federation of Teachers (AFT), the United Federation of Teachers and Microsoft. It is between the National Academy for AI Instruction, an educator-led training body announced by the AFT in 2025, and Microsoft, with blank signature spaces for other providers. Districts may incorporate its terms into their own Microsoft contracts. The source says a district could terminate a contract and seek damages for a breach; the agreement does not automatically create a legal duty for every school or AI provider.
For covered education products, the agreement prohibits using student, educator or administrator data—including prompts, uploaded files, behavioral signals and metadata—to train or improve AI models. It also bars specified tracking, including continuous passive attention monitoring and emotional inference, and prohibits companion-style features designed to foster emotional attachment. A narrow exception allows data processing to detect specified safety risks, such as threats of violence or self-harm; that data cannot then be used for general model improvement. The agreement also says covered systems cannot make decisions about students or teachers without meaningful human review or prior approval.
The agreement has limits and exceptions. It covers AI products designed and marketed for education, while general-purpose productivity and workplace tools are expressly excluded. Some prohibited monitoring may be allowed with a school customer’s explicit written approval. The text also calls for reporting breaches within 72 hours, deleting data from active systems within 180 days, and keeping data in the EU or North America. Microsoft, the agreement and press accounts describe differently whether protections apply automatically or only when districts request or contract for them; the source does not resolve that discrepancy.
Infographic 01 · The lines drawn this autumn
Seven weeks, one direction: a human stays in charge
From the school contract to state laws: what was signed when, and what the classroom standard actually requires.
- 2 Feb 2025EU: emotion inference banned at work and schoolAI Act Art. 5(1)(f), except medical or safety reasons
- 9 Sep 2026AFT, UFT and Microsoft announce the school standardA contract districts can add to their Microsoft agreements
- 30 Sep 2026California signs AB 1883, SB 947, SB 951, AB 1979Emotion AI at work, robot bosses, AI layoff notices, AI in clinics
- 28 Oct 2026New York signs its chatbot law for minorsS.9051B/A.10379C: no unsafe companion features for under-18s
- 1 Nov 2026Microsoft makes the standard available to all US districtsPer Microsoft’s fact sheet; districts “can add them”
- 1 Jan 2027Most California rules take effectAB 1883, AB 1979, SB 951
- 1 Jul 2027SB 947 becomes operativeHuman corroboration before AI-driven discipline or firing
What Microsoft’s classroom AI may not do
- Learn from students or teachers. No training on their data; de-identification is “not a loophole”.
- Track them. No keystroke logging, attention monitoring, long-term profiling or biometrics.
- Befriend them. No companion features designed to foster emotional attachment.
- Decide alone. No decisions without meaningful human review or prior approval.
Where it is narrower
- Scope: education products only. General-purpose productivity and workplace tools are excluded by definition.
- Opt-in or automatic: Microsoft says districts “can add” the terms; the agreement says “upon request within 90 days”.
- Exceptions: a narrow safety exception for training; the no-tracking list allows written approval by the school.
“Anything less than legally enforceable provisions is simply a wish list.”Randi Weingarten, AFT President, 9 September 2026
Sources: AFT/UFT/Microsoft release, 9 Sep 2026 · Standard MOA v1.0 · Microsoft fact sheet · Governor of California, 30 Sep 2026 · Governor of New York, 28 Oct 2026 · EU AI Act, Art. 5(1)(f). Effective dates for California follow the chaptered texts and the state’s default start date for regular-session bills.
Powered by Thorsten Meyer AI
Where Human Review Becomes a Requirement
The measures address a practical question for schools, employers and families: which decisions should remain accountable to a person when AI is used to assess, recommend or interact with people. In schools, contractual terms could give districts a procurement tool to demand limits from a supplier, even though the agreement is not legislation. The distinction matters: a district’s remedies would depend on its contract, rather than on the agreement automatically binding all vendors.
California’s laws extend the debate into employment and health care, while New York’s measure concerns chatbots interacting with children. Together, the developments make human oversight, sensitive-data collection and emotionally persuasive design concrete compliance issues. They do not establish a general prohibition on AI. For organizations, the immediate work is to identify where tools infer mood or attention, where software informs discipline or termination, and what human review actually occurs.
Products Worth Considering
Apple MacBook Air 13-inch with M5 chip, 16GB memory, 1TB SSD, and 12MP Center Stage camera (Midnight)
DJI Osmo Mobile 8 Standard Combo Gimbal Stabilizer for Phone
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
As an affiliate, we earn on qualifying purchases.
Three Different Routes to Oversight
California Governor Gavin Newsom signed four measures on September 30, 2026. The source describes AB 1883 as prohibiting AI that reads workers’ emotions or collects neural data; SB 947 as barring discipline or firing decided by an algorithm alone; AB 1979 as preventing AI from performing health-care functions reserved for licensed professionals; and SB 951 as requiring layoff notices to disclose when AI replaced jobs. Most are scheduled to take effect January 1, 2027; SB 947 is scheduled for July 1, 2027.
New York Governor Kathy Hochul signed a chatbot law on October 28. As described in the source, it restricts unsafe companion features for minors and unverified users, including chatbots that present themselves as human or encourage emotional dependency. These state laws are distinct from Microsoft’s voluntary-to-contract school framework: their scope and enforcement derive from statutory text, whereas the school provisions depend on procurement and contract terms.
The AFT’s Randi Weingarten argued that enforceability is central, saying, “anything less than legally enforceable provisions is simply a wish list.” Microsoft President Brad Smith called the agreement a standard that “sets a high bar for child privacy and AI safety.” Those are proponents’ characterizations; the agreement’s practical reach will depend on adoption by districts and the terms in their contracts.
““Anything less than legally enforceable provisions is simply a wish list.””
— Randi Weingarten, president of the American Federation of Teachers
Adoption and Scope Remain Open
It is not clear how many districts will adopt the agreement, when they would add its terms to contracts, or whether other AI providers will sign. The source also reports inconsistent descriptions of whether Microsoft protections apply by default or only after a district requests them. That point affects how schools should understand the offer and should be checked against Microsoft’s contract documents and each district’s agreement.
The source does not provide the full statutory text or explain enforcement mechanisms for the California and New York measures. Their precise application will depend on the final laws, implementing guidance and specific circumstances. The school agreement’s exceptions—including written school approval for some monitoring—also mean that its restrictions are not absolute in every case.
District Contracts and Effective Dates
School districts can review Microsoft’s offer and decide whether to incorporate its provisions into procurement contracts. They will need to check which products are covered, how human review is defined, who can authorize exceptions, and how breach reporting and data deletion are documented. The agreement also calls for agent features that act outside a provider’s systems to be disabled for students unless an administrator enables them.
California’s measures are scheduled to begin taking effect in January 2027, with SB 947 following in July. Organizations affected by those laws will need to review their tools and decision processes before the relevant dates. Further details on enforcement, implementation and district adoption will determine how much these commitments change day-to-day AI use.
Key Questions
Is Microsoft’s school AI agreement a law?
No. It is a memorandum intended for incorporation into contracts between Microsoft and participating school districts. Its enforceability would depend on the resulting contract, not a nationwide statute.
What does the agreement say about student data?
For covered education products, it prohibits using student and educator data to train or improve AI models, with a limited safety-processing exception that cannot be used for general model improvement.
Does the agreement cover every Microsoft AI product?
No. The source says it covers products designed and marketed for education and expressly excludes general-purpose productivity and workplace tools.
When do California’s AI laws take effect?
According to the source, most of the four measures take effect January 1, 2027. SB 947, concerning algorithm-only decisions on discipline or firing, is scheduled to take effect July 1, 2027.
What remains uncertain about the school agreement?
Its adoption by districts, whether protections apply automatically or only after a request, and how its exceptions will work in practice remain unclear from the available source material.
Source: ThorstenMeyerAI.com
